Developing Smart Emergency Applications with Multi-Agent Systems

Introduction

Multi-agent systems are one of the most interesting areas in software research and they have been importantly contributing to the development of the theory and the practice of complex distributed systems (see, e.g., Bordini et al., 2005). In particular, multi-agent systems have shown the potential to meet critical needs in high-speed, mission-critical, content-rich, distributed information systems where mutual interdependencies, dynamic environments, uncertainty, and sophisticated control play a singular role (Gasser, 2001).

Agent and multi-agent system are buzzwords that found their way into a number of technologies and they have been largely used, e.g., in Artificial Intelligence, Databases, Operating Systems and Computer Networks. Although there is no such thing as an accepted definition of an agent (see, e.g., Genesereth & Ketchpel, 1994; Wooldridge & Jennings, 1995), all proposed definitions agree that an agent is essentially an autonomous software entity that provides an interoperable interface and that behaves like a rational actor working on behalf of some client in pursuit of its own agenda. Agents are designed to operate in dynamic and uncertain environments, taking complex decisions at run-time, and the learning capabilities of some agents make them capable to improve their performances over time, thus avoiding repeated negative conditions and persisting on successful behaviours.

Even if a complex system can be realized in terms of a solitary agent working within its environment–that may or may not comprise users–usually, agent-based systems are made of multiple, interacting agents: agent-based systems are normally multi-agent systems. Multi-agent systems are generally considered an appropriate means for modelling complex, distributed systems, even if such a multiplicity naturally introduces the possibility of having different agents with potentially conflicting goals. Agents in a multi-agent system may decide to cooperate for mutual benefit, or they may compete to serve their own interests. They may take advantage of their social ability to exhibit flexible coordination behaviours that make them able to cooperate in the achievement of shared goals or to compete on the acquisition of resources and tasks. Finally, agents in a multi-agent system have the ability of coordinating their behaviours into coherent global actions. Coordination among agents is handled by means of a variety of approaches including negotiation (Jennings, 2001), organizational structuring (Horling & Lesser, 2005) and multi-agent planning (Durfee, 1999).

The very fact of characterizing agents in terms of the properties that they exhibit rather than in terms of a crude and often inapplicable definition, allows a plethora of software systems to be considered as first-class multi-agent systems. This makes multi-agent systems much more than a single technology supporting the realization of complex distributed systems. Multi-agent systems are abstractions capable of capturing the essence of many software systems at different levels of detail. In particular, agents and multi-agent systems are often considered the highest system level (Jennings, 2000) of today computing systems and they are meant to provide a truly novel level of abstraction in the analysis, design and implementation of complex software systems (Bergenti & Huhns, 2004). This is the reason why we can correctly account many recent software systems as multi-agent systems, even if no agent-based development technology were adopted in their realization. The agent-based nature of a system comes from the characteristics of its components and of the interactions among them, rather than from a hypothetical “agent based” label attached on the box of the development tool adopted for their realization. Notably, multi-agent systems are often developed using technologies that have no built-in notion of agent.